检索结果-内蒙古大学图书馆

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Suntae Hwang Seonghyeon Kang Kyungsu Kim Semin Ahn Kyogu Lee Department of Intelligence and Information Music and Audio Research Group (MARG) Seoul National University Department of Mechanical Engineering Seoul National University Interdisciplinary Program in Artificial Intelligence Seoul National University Artificial Intelligence Institute Seoul National University

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

Drum one-shot samples are crucial for music production, particularly in sound design and electronic music. This paper introduces Drum One-Shot Extraction, a task in which the goal is to extract drum one-shots that are present in the music mixture. To facilitate this, we propose the Random Mixture One-shot Dataset (RMOD), comprising large-scale, randomly arranged music mixtures paired with corresponding drum one-shot samples. Our proposed model, Drum One-Shot Extractor (DOSE), leverages neural audio codec language models for end-to-end extraction, bypassing traditional source separation steps. Additionally, we introduce a novel onset loss, designed to encourage accurate prediction of the initial transient of drum one-shots, which is essential for capturing timbral characteristics. We compare this approach against a source separation-based extraction method as a baseline. The results, evaluated using Fréchet Audio Distance (FAD) and Multi-Scale Spectral loss (MSS), demonstrate that DOSE, enhanced with onset loss, outperforms the baseline, providing more accurate and higher-quality drum one-shots from music mixtures. The code, model checkpoint, and audio examples are available at https://***/HSUNEH/DOSE

关键词： Training Measurement Accuracy Source separation Instruments Production Data models Electronic music Transient analysis Speech processing

来源：评论

学校读者我要写书评

暂无评论

Graph neural networks for classification and error detection in 2D architectural detail drawings

引用

AUTOMATION IN CONSTRUCTION 2025年 170卷

作者： Ko, Jaechang Lee, Donghyuk Texas A&M Univ Dept Architecture College Stn TX 77840 USA Seoul Natl Univ Interdisciplinary Program Artificial Intelligence 1 Gwanak Ro Seoul 08826 South Korea

The assessment and classification of architectural sectional drawings is critical in the architecture, engineering, and construction (AEC) field, where the accurate representation of complex structures and the extraction of meaningful patterns are key challenges. This paper established a framework for standardizing different forms of architectural drawings into a consistent graph format, and evaluated different Graph Neural Networks (GNNs) architectures, pooling methods, node features, and masking techniques. This paper demonstrates that GNNs can be practically applied in the design and review process, particularly for categorizing details and detecting errors in architectural drawings. The potential for visual explanations of model decisions using Explainable AI (XAI) is also explored to enhance the reliability and user understanding of AI models in architecture. This paper highlights the potential of GNNs in architectural data analysis and outlines the challenges and future directions for broader application in the AEC field.

关键词： Architectural detail drawings Graph neural networks (GNN) Drawing classification Error detection Explainable AI (XAI)

来源：评论

学校读者我要写书评

暂无评论

MathReader : Text-to-Speech for Mathematical Documents

MathReader : Text-to-Speech for Mathematical Documents

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Sieun Hyeon Kyudan Jung Nam-Joon Kim Hyun Gon Ryu Jaeyoung Do Department of Electrical and Computer Engineering Seoul National University Department of Mathematics Chung-Ang University NVIDIA Interdisciplinary Program in Artificial Intelligence Seoul National University

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

TTS (Text-to-Speech) document reader from Microsoft, Adobe, Apple, and OpenAI have been serviced worldwide. They provide relatively good TTS results for general plain text, but sometimes skip contents or provide unsatisfactory results for mathematical expressions. This is because most modern academic papers are written in LaTeX, and when LaTeX formulas are compiled, they are rendered as distinctive text forms within the document. However, traditional TTS document readers output only the text as it is recognized, without considering the mathematical meaning of the formulas. To address this issue, we propose MathReader, which effectively integrates OCR, a fine-tuned T5 model, and TTS. MathReader demonstrated a lower Word Error Rate (WER) than existing TTS document readers, such as Microsoft Edge and Adobe Acrobat, when processing documents containing mathematical formulas. MathReader reduced the WER from 0.510 to 0.281 compared to Microsoft Edge, and from 0.617 to 0.281 compared to Adobe Acrobat. This will significantly contribute to alleviating the inconvenience faced by users who want to listen to documents, especially those who are visually impaired. The code is available at https://***/hyeonsieun/MathReader.

关键词： Text recognition Error analysis Pipelines Optical character recognition Graphics processing units Signal processing Real-time systems Mathematical models Text to speech Speech processing

来源：评论

学校读者我要写书评

暂无评论

Learning 3D Scene Analogies with Neural Contextual Scene Maps

arXiv

引用

arXiv 2025年

作者： Kim, Junho Bae, Gwangtak Lee, Eun Sun Kim, Young Min Dept. of Electrical and Computer Engineering Seoul National University Korea Republic of Interdisciplinary Program in Artificial Intelligence and INMC Seoul National University Korea Republic of

Understanding scene contexts is crucial for machines to perform tasks and adapt prior knowledge in unseen or noisy 3D environments. As data-driven learning is intractable to comprehensively encapsulate diverse ranges of layouts and open spaces, we propose teaching machines to identify relational commonalities in 3D spaces. Instead of focusing on point-wise or object-wise representations, we introduce 3D scene analogies, which are smooth maps between 3D scene regions that align spatial relationships. Unlike well-studied single instance-level maps, these scene-level maps smoothly link large scene regions, potentially enabling unique applications in trajectory transfer in AR/VR, long demonstration transfer for imitation learning, and context-aware object rearrangement. To find 3D scene analogies, we propose neural contextual scene maps, which extract descriptor fields summarizing semantic and geometric contexts, and holistically align them in a coarse-to-fine manner for map estimation. This approach reduces reliance on individual feature points, making it robust to input noise or shape variations. Experiments demonstrate the effectiveness of our approach in identifying scene analogies and transferring trajectories or object placements in diverse indoor scenes, indicating its potential for robotics and AR/VR applications. © 2025, CC BY.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

Fate-tox: fragment attention transformer for E(3)-equivariant multi-organ toxicity prediction

引用

Journal of Cheminformatics 2025年第1期17卷 1-16页

作者： Ha, Sumin Bang, Dongmin Kim, Sun Interdisciplinary Program in Artificial Intelligence Seoul National University Seoul 08826 South Korea Interdisciplinary Program in Bioinformatics Seoul National University Seoul 08826 South Korea AIGENDRUG Co. Ltd. Seoul 08758 South Korea Department of Computer Science and Engineering Seoul National University Seoul 08826 South Korea

Toxicity is a critical hurdle in drug development, often causing the late-stage failure of promising compounds. Existing computational prediction models often focus on single-organ toxicity. However, avoiding toxicity of an organ, such as reducing gastrointestinal side effects, may inadvertently lead to toxicity in another organ, as seen in the real case of rofecoxib, which was withdrawn due to increased cardiovascular risks. Thus, simultaneous prediction of multi-organ toxicity is a desirable but challenging task. The main challenges are (1) the variability of substructures that contribute to toxicity of different organs, (2) insufficient power of molecular representations in diverse perspectives, and (3) explainability of prediction results especially in terms of substructures or potential toxicophores. To address these challenges with multiple strategies, we developed FATE-Tox, a novel multi-view deep learning framework for multi-organ toxicity prediction. For variability of substructures, we used three fragmentation methods such as BRICS, Bemis-Murcko scaffolds, and RDKit Functional Groups to formulate fragment-level graphs so that diverse substructures can be used to identify toxicity for different organs. For insufficient power of molecular representations, we used molecular representations in both 2D and 3D perspectives. For explainability, our fragment attention transformer identifies potential 3D toxicophores using attention coefficients. Scientific contribution: Our framework achieved significant improvements in prediction performance, with up to 3.01% gains over prior baseline methods on toxicity benchmark datasets from MoleculeNet (BBBP, SIDER, ClinTox) and TDC (DILI, Skin Reaction, Carcinogens, and hERG), while the multi-task learning approach further enhanced performance by up to 1.44% compared to the single-task learning framework that had already surpassed these baselines. Additionally, attention visualization aligning with literature contributes to

关键词： Graph neural network Multi-task learning Toxicity prediction Transformer

来源：评论

学校读者我要写书评

暂无评论

Cinema Multiverse Lounge: Enhancing Film Appreciation via Multi-Agent Conversations 25

Cinema Multiverse Lounge: Enhancing Film Appreciation via Mu...

引用

Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems

作者： Jeongwoo Ryu Kyusik Kim Dongseok Heo Hyungwoo Song Changhoon Oh Bongwon Suh Department of Intelligence and Information Seoul National University Seoul Republic of Korea Interdisciplinary Program in Artificial Intelligence Seoul National University Seoul Republic of Korea Graduate School of Information Yonsei University Seoul Republic of Korea

来源：评论

学校读者我要写书评

暂无评论

Development of EEG-Based Optimal Biomarkers for Decoding Fatigue during Motor Imagery Tasks

Development of EEG-Based Optimal Biomarkers for Decoding Fat...

引用

International Winter Workshop on Brain-Computer Interface (BCI)

作者： Jun-Seok Lee Seung-Won Lee Sang-Rok Kim Han-Jeong Hwang Dept. Electronics and Information Engineering Korea University Sejong Republic of Korea Interdisciplinary Graduate Program for Artificial Intelligence Smart Convergence Technology Korea University Sejong Republic of Korea

ISBN: (数字)9798331521929

ISBN: (纸本)9798331521936

This study aimed to identify electroencephalography (EEG)-based neurophysiological biomarkers for monitoring mental fatigue during motor imagery (MI) tasks. We used EEG data recorded from 29 healthy adults during MI tasks. For the preprocessing of the EEG data, we sequentially performed down-sampling into 200 Hz, independent component analysis (ICA), band-pass filtering between 0.5 and 50 Hz, and common average reference (CAR). The preprocessed EEG data were analyzed using a fast Fourier transform (FFT) to calculate spectral power in the δ (0.5-3 Hz), θ (4–7 Hz), $a$ (8–13 Hz), and β (13–30 Hz) frequency bands. Subsequently, seventeen fatigue-related metrics were calculated using individual spectral powers and their combinations. These fatigue-related metrics were averaged for the frontal lobe (F3 and F4), central lobe (FCC3h and FCC4h), parietal lobe (P3 and P4), and occipital lobe (POO1 and POO2). The statistical significance of the average fatigue metrics was evaluated through Pearson correlation analysis with the sequence of trials. As a result, we observed strong positive correlations (r < 0.9, p < 0.001) between beta power in the frontal and central regions and the sequence of trials. These findings suggest that central and frontal beta activity can serve as reliable biomarkers for estimating fatigue during MI tasks, supporting their potential for real-time fatigue monitoring.

关键词： Measurement Correlation Fast Fourier transforms Biomarkers Fatigue Motors Electroencephalography Decoding Reliability Monitoring

来源：评论

学校读者我要写书评

暂无评论

GraphT5: Unified Molecular Graph-Language Modeling via Multi-Modal Cross-Token Attention

arXiv

引用

arXiv 2025年

作者： Kim, Sangyeup Kim, Nayeon Piao, Yinhua Kim, Sun Computer Science and Engineering Seoul National University Korea Republic of Interdisciplinary Program in Artificial Intelligence Seoul National University Korea Republic of AIGENDRUG Co. Ltd Korea Republic of

Molecular language modeling tasks such as molecule captioning have been recognized for their potential to further understand molecular properties that can aid drug discovery or material synthesis based on chemical reactions. Unlike the common use of molecule graphs in predicting molecular properties, most methods in molecular language modeling rely heavily on SMILES sequences. This preference is because the task involves generating a sequence of multiple tokens using transformer-based models. Therefore, a main challenge is determining how to integrate graph data, which contains structural and spatial information about molecules, with text data. In addition, simply using both 1D SMILES text and 2D graph as inputs without addressing how they align and represent the molecule structure in different modalities makes it challenging to fully utilize structural knowledge about molecules. To this end, we propose GraphT5, a multi-modal framework that integrates 1D SMILES text and 2D graph representations of molecules for molecular language modeling. Specifically, we introduce a novel cross-token attention module in GraphT5 to bridge the gap arising from the fundamental differences between the two modalities of molecule representations. Cross-token attention exploits implicit information between SMILES and graphs of molecules, resulting from their interactions at a fine-grained token level that benefits molecular language modeling. Extensive experiments including molecule captioning, IUPAC name prediction tasks, and case studies show that our GraphT5 outperforms the latest baseline approaches, which validates the effectiveness of our GraphT5 in sufficiently utilizing 1D SMILES text and 2D graph representations. © 2025, CC BY-NC-ND.

关键词： Unified Modeling Language

来源：评论

学校读者我要写书评

暂无评论

Increased local DNA methylation disorder in AMLs with DNMT3A-destabilizing variants and its clinical implication

引用

NATURE COMMUNICATIONS 2025年第1期16卷 1-15页

作者： Lee, Dohoon Koo, Bonil Kim, Seokhyeon Byun, Jamin Hong, Junshik Shin, Dong-Yeop Sun, Choong-Hyun Kim, Jaesung Song, Ji-Joon Jaiswal, Siddhartha Yoon, Sung-Soo Kim, Sun Koh, Youngil Seoul Natl Univ Bioinformat Inst Seoul South Korea Seoul Natl Univ BK21 FOUR Intelligence Comp Seoul South Korea Seoul Natl Univ Interdisciplinary Program Bioinformat Seoul South Korea AIGENDRUG Co Ltd Seoul 08826 South Korea Seoul Natl Univ Canc Res Inst Seoul South Korea Seoul Natl Univ Hosp Dept Internal Med Seoul South Korea Seoul Natl Univ Hosp Ctr Med Innovat Seoul South Korea Genome Opinion Inc Seoul South Korea Korea Adv Inst Sci & Technol KAIST Dept Biol Sci KI BioCentury Daejeon South Korea Stanford Univ Dept Pathol Stanford CA USA Seoul Natl Univ Interdisciplinary Program Artificial Intelligence Seoul South Korea Seoul Natl Univ Dept Comp Sci & Engn Seoul South Korea Mogam Inst Biomed Res Yongin South Korea

The mechanistic link between the complex mutational landscape of de novo methyltransferase DNMT3A and the pathology of acute myeloid leukemia (AML) has not been clearly elucidated so far. Motivated by a recent discovery of the significance of DNMT3A-destabilizing mutations (DNMT3AINS) in AML, we here investigate the common characteristics of DNMT3AINS AML methylomes through computational analyses. We present that methylomes of DNMT3AINS AMLs are considerably different from those of DNMT3AR882 AMLs in that they exhibit increased intratumor DNA methylation heterogeneity in bivalent chromatin domains. This epigenetic heterogeneity was associated with the transcriptional variability of developmental and membrane-associated factors shaping stem cell niche, and also was a predictor of the response of AML cells to hypomethylating agents, implying that the survival of AML cells depends on stochastic DNA methylations at bivalent domains. Altogether, our work provides a novel mechanistic model suggesting the genomic origin of the aberrant epigenomic heterogeneity in disease conditions.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Correction: Learning distribution-free anchored linear structural equation models in the presence of measurement error

引用

Journal of the Korean Statistical Society 2025年 1-2页

作者： Chung, Junhyoung Ahn, Youngmin Shin, Donguk Park, Gunwoong Department of Statistics Seoul National University Seoul Republic of Korea Institute for Data Innovation in Science Seoul National University Donguk Shin Seoul Republic of Korea Interdisciplinary Program in Artificial Intelligence Seoul National University Seoul Republic of Korea

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：